Issues in Advanced Lithography
05/01/1997
Issues in advanced lithography
Katherine Derbyshire, Chief Technical Editor
For many years, optical lithography has been the engine driving Moore`s Law. Each device generation has brought new fears that, this time, optical lithography will no longer suffice, and each generation has proved those fears unfounded. Mercury lamps have given way to excimer lasers, proximity masks have been replaced by 1?, then 4? and 5? projection reticles. Complex resist systems now rely on chemical amplification and antireflective coatings. This article looks at some of the forces driving the latest lithography transition, and some of the questions that remain to be answered as Moore`s Law faces the dawn of a new century.
Constant feature shrinks are so much a part of the semiconductor industry that we sometimes take them for granted. It`s worth remembering the underlying economic forces. The processing cost of a wafer is more or less independent of the number of die on the wafer. So, the higher the circuit density, the lower the cost/packaged device. In the DRAM market, customers are primarily concerned with cost/bit. More chips/wafer means more bits, for a lower cost/bit and higher profit to the manufacturer. This model assumes that the increased cost of more advanced fabs will be less than the gains from economies of scale. So far, this assumption has been true. Memory companies have driven technology evolution for many years because their business model has required faster cost cutting than other industry sectors.
For logic (including microprocessors and microcontrollers), the economics are more complicated. More chips/wafer still means more profit, but logic customers are also willing to pay a premium for higher performance. Higher-density circuits run faster, because, in simplistic terms, electrons don`t have to travel as far in order to switch a gate or transmit signals. The channel length dominates the performance of individual transistors, so this "critical layer" requires the tightest possible critical dimension (CD) control. Motorola`s John Sturtevant estimates that 1 nm in channel CD variation is equivalent to 1 MHz chip-speed variation, and is thus worth about $7.50 in the selling price of each chip. He therefore claims that leading edge products require ?8% 3-s CD control [1].
With that strong financial incentive, logic makers are accelerating their technology transitions. The 1994 SIA Roadmap called for 0.25-?m lithography to be used at critical levels in 1998, and required only ?10% CD variation (see table) [2]. Instead, according to John Wiesner, senior VP of engineering, Nikon expects to ship more than 200, 0.25-?m tools in 1997. Observers also say that the 0.25-?m generation is using 0.20-?m gates. Later generations will be here faster than predicted as well. Logic, not memory, now appears to be driving technology evolution.
Against this background, a debate rages about lithography for 0.18-?m features and smaller, now expected to be needed in production by 2000 or 2001. The three exposure alternatives appear to be 248 nm, 193 nm, and x-ray.
Extending DUV: 248 nm
Deep-ultraviolet (DUV) lithography uses a 248-nm, KrF excimer laser, rather than a mercury lamp, for exposure. Though Advanced Micro Devices showed that i-line can print 0.25-?m features, most new fabs are using DUV [3]. Printing 0.18-?m (180-nm) features with 248-nm light is difficult because the features are substantially smaller than the exposure wavelength. The Rayleigh equation states that resolution at a given wavelength can be improved with large NA optics, but large numerical apertures enforce a very narrow DOF. With conventional masks and illumination, the process window becomes unworkably narrow.
Masks and exposure strategies. As it turns out, though, several optical tricks can expand the process latitude. Off-axis illumination (OAI) was the first of these to be used in production. Off-axis techniques manipulate the light source to shift the angle of incidence of the zero-order beam (Fig. 1) [4]. OAI thus maintains high contrast and acceptable DOF in the aerial image at high resolution, but does reduce the amount of light reaching the wafer plane.
Figure 1. Optical paths in steppers for a) conventional three-beam illumination and b) off-axis illumination. [4]
As researchers at Samsung found, unbalanced OAI intensity deforms the exposed pattern. This effect, a challenge for system designers, is especially severe for small features near the resolution limit [5]. Even when intensity is balanced, OAI also tends to degrade the boundaries of circuit patterns. Optical proximity correction (OPC), the second important optical trick, predistorts masks with extra subresolution features so that these boundary features will print as desired. OPC can also correct for iso-dense bias -the difference in printed linewidth between an isolated line and a line in a dense array - due to both lithography and later resist and etch processing. The corrective features may simply extend or widen lines (1-D OPC), or may use notches and serifs to eliminate corner rounding (2-D OPC). It is not yet clear which features will provide the most effective correction with the least added complexity.
Phase-shift masks (PSM), another correction for these problems, use destructive interference to improve resolution and DOF. Several different PSM strategies are available, but alternating-aperture (AAPSM) and chromeless phase-edge designs appear to provide the most improvement (Fig. 2). AAPSMs significantly improve periodic structures, but nonperiodic features have been difficult to implement. Unwanted phase-shift lines in nonperiodic structures print as dark image features unless erased or modified by subsequent exposure. AAPSMs are also difficult to fabricate since they use two independent mask layers. One of these, the phase-shift layer, must be etched to a precise depth [6].
Figure 2. Varieties of photomasks with optical amplitude profiles in the object plane; a) conventional transmission mask, b) alternating-aperture mask, and c) chromeless phase-edge mask. [6]
Both OPC and PSM use the mask to correct for process characteristics. They therefore require a radical change in the relationship among designers, maskmakers, and process engineers. Right now, the relationship is linear, design-to-mask-to-process, with little or no communication. Designers don`t like design changes at the process development stage, and process engineers expect the mask to behave in a consistent way as the process changes [7]. This approach won`t work with advanced masks, as Motorola`s Kevin Lucas explained [8]:
First, since subresolution features are used, it is very difficult to verify a mask without actually printing wafers from it. For example, OPC masks deliberately violate conventional design rules, so normal design-rule checking will reject good masks. Process expertise must be incorporated at the mask and design stages.
Second, advanced masks require accurate reproduction of features smaller than the nominal mask CD. Mask processing must thus advance even more rapidly than wafer processing. The necessary research and development expenditures could be substantial, participants in a BACUS panel discussion at the recent SPIE Microlithography symposium agreed, yet maskmaking has historically been a low-margin business. Semiconductor companies may need to help support mask-process research.
Third, and perhaps most problematic, advanced masks are generally designed to correct for a specific process condition. The mask pattern may need to change during process optimization. In manufacturing, an unstable process will undermine the OPC scheme. The mask shop and designers must be involved in the process feedback loop.
All of these factors require much closer cooperation among mask houses, designers, and process engineers than has previously occurred. Still, according to phase-shift mask researcher Marc Levenson, the barriers are primarily cultural, not technical [9].
Resists. Resist chemistry presents unique challenges as well. Conventional novolac/diazoquinone resists have low sensitivity to DUV light, yet resolution enhancement techniques like OAI deliver less light to the wafer plane. To achieve acceptable photospeed, DUV lithography normally uses chemically amplified (CA) resists. These systems, IBM`s Hiroshi Ito explains, contain a photoacid generator compound that decomposes when exposed to light. During postexposure bake (PEB), the acid decomposition product induces a cascade of chemical transformations, changing the solubility of exposed regions. Early formulations tended to form an insoluble skin and are generally sensitive to atmospheric contamination between exposure and PEB; but newer chemistries seem to be more robust [10]. Still, resist process control is a key user concern. Researchers at the University of Texas, Austin, in collaboration with AZ Photoresist and Hoechst-Celanese, have developed a non-CA, diazocoumarin-based resist optimized for 248-nm exposure [11].
Reflected light from the wafer can expose resist, too, sometimes creating standing wave "ripples" on the edges of features. A variety of top-and-bottom antireflective coatings can prevent this effect. Resists can also be optimized for isolated or dense features. Bill Krisa of Texas Instruments predicts that 0.25-?m lithography will use three or four different resist formulations [12].
Difficult though 248-nm exposure is, 193-nm and x-ray pose their own problems for engineers. The accelerated transition to 0.18 ?m will probably mean that neither of these technologies is mature enough to be used, but one or both will still be needed for later generations.
Still optics, sort of: 193 nm
Most signs point to 193-nm (ArF laser) lithography as the successor to DUV. This wavelength is still larger than the feature size at 0.18 ?m or less, so versions of the optical tricks discussed above will be necessary. ArF lithography also faces some unique challenges of its own.
Most crystalline lens materials, like quartz and CaF2, absorb strongly at 193 nm. Some, like CaF2, are also difficult to polish. Fused silica is, according to William Oldham of the University of California, Berkeley, currently the only choice for large refractive elements. The high-intensity excimer laser beam tends to densify, or "compact," fused silica lenses, though, changing the refractive index and eventually rendering the lens unusable. Lifetime estimates range from a few months to several years, depending on lens design and silica quality. Even in the worst case, reasonable lifetimes can be achieved with purely refractive, fused silica lenses by lowering the peak optical intensity. More sensitive resists are one alternative, reducing optical intensities while maintaining high throughput. Exclusive use of dark-field masks would lower the energy density in most of the optical column, but not at the image plane. These masks would require both positive and negative tone resists [13]. According to SVG Lithography`s VP for technology and marketing, John Shamaly, catadioptric lens designs, containing both reflective and refractive elements, can substantially eliminate compaction concerns.
Conventional resists absorb strongly at 193 nm. Investigations of new chemistries are still at a relatively early stage, but most researchers seem to favor methacrylate-based polymers. Shipley`s C.R. Szmanda reports that early problems with poor adhesion and poor dissolution control in standard TMAH developer can be addressed with added monomers [14]. Other resist chemistries, notably cyclo-olefins [15], have been investigated as well. It is not clear which option will prove most manufacturable. At the moment, resists seem to be the biggest question mark in 193-nm development.
Always the bridesmaid: X-ray
Given the limitations of 193-nm lithography, x-ray (about 8-? wavelength) would seem to be an obvious solution and, in fact, researchers have been investigating it for several decades.
Current x-ray exposure systems use simple proximity exposure with 1? masks, avoiding complex lens designs altogether. Yet, that approach is part of the problem - proximity printing requires tight control of a narrow (about 10 ?m) gap between the mask and the wafer. As x-ray radiation heats the mask, distortion may change the gap. Thermal stability of Si/Au masks has been iffy at best, but SiC/Ta masks (see Technology News, "IBM, Motorola, Japan firms agree on common mask for x-ray lithography") should improve matters. Proximity printing also risks contamination and damage to both the wafer and the mask.
Another difficulty arises because x-ray masks are 1:1. The mask feature size is much smaller than in equivalent reduction systems. Direct-write e-beam can produce 0.18-?m features, barely, but throughput is low. So far, no one has attempted volume production of x-ray masks. Observers agree that masks are the limiting piece of the x-ray puzzle.
X-ray development also lacks economical exposure sources. While synchrotron rings are mature and reliable, and provide a stable, well-collimated beam, the ring alone costs $15-20 million, plus the cost of beam lines and enclosures. One such tool can support up to 16 steppers, bringing volume production costs into line with DUV, but fabs are reluctant to rely on one source for all critical layer lithography. Worse, the large start-up cost discourages small-scale development. Despite ongoing research efforts [16], a commercial x-ray point source remains elusive.
Perhaps the most serious barrier to x-ray lithography, though, is cultural. As Nikon`s John Wiesner explains, "1? x-ray people have done a yeoman job of positioning themselves to be a backup technology." The semiconductor community, however, prefers an optical solution and will use one if available.
Way out there: The far future
Beyond these alternatives, the lithography picture becomes even more confused. Extreme ultraviolet (EUV, about 13 nm), like x-ray, would offer a substantially greater resolution budget than 193 nm. EUV lenses use strictly reflective elements and, according to Tropel president John Bruning, require coatings as precise as those used in space telescopes [17]. (It is worth noting here that SVGL recently entered an optical collaboration with NASA, see Technology News, "SVG Litho, NASA enter optical collaboration.") The most serious obstacle is the light source, though none of the several proposed EUV sources has yet been built. Direct-write and projection e-beam exposure have garnered strong interest from SEMATECH, which has funded the SCALPEL (scattering with angular limitation projection electron-beam lithography) program [18], but throughput is a concern.
Conclusion
For the near term, the industry consensus seems to be coalescing around 248-nm exposure for 0.18-?m features. The most likely successor seems to be 193 nm, but its market entry point is unclear. X-ray and other technologies could still grab the industry`s attention if 193-nm resist development falters.
References
1. J. Sturtevant et al., "CD Control Challenges for Sub-0.25 ?m Patterning," SEMATECH DUV Lithography Workshop, Austin, TX, Oct. 16-18, 1996.
2. Semiconductor Industry Association, The National Technology Roadmap for Semiconductors, 1994.
3. P. Ackmann et al., "Phase Shifting and Optical Proximity Corrections to Improve CD Control on Logic Devices in Manufacturing for Sub-0.35 ?m I-line," SPIE`s 22nd International Symposium on Microlithography, Paper 3051-07, March 9-14, 1997.
4. J. Mulkens et al., "High-throughput Wafer Stepper with Adjustable Illumination Modes," Solid State Technology, pp. 193-199, July 1996.
5. J-H. Kim et al., "Pattern Deformation Induced from Intensity-unbalanced Off-axis Illumination," SPIE`s 22nd International Symposium on Microlithography, Paper 3051-02, March 9-14, 1997.
6. M.D. Levenson, "Extending Optical Lithography to the Gigabit Era," Microlithography World, pp. 5-13, Autumn 1994.
7. F. Shellenberg, "The Litho/Design Workshop," Solid State Technology, pp. 56-61, Oct. 1996.
8. K. Lucas et al., "Practical Process Corrections for Advanced Logic Designs," SEMATECH DUV Lithography Workshop, Austin, TX, Oct. 16-18, 1996.
9. M.D. Levenson, "Wavefront Engineering from 500 nm to 100 nm CD," SPIE`s 22nd International Symposium on Microlithography, plenary address, March 9-14, 1997.
10. H. Ito, "Deep-UV Resists: Evolution and Status," Solid State Technology, pp. 164-173, July 1996.
11. C.G. Willson et al., "A Non-chemically Amplified 248 nm Resist," SPIE`s 22nd International Symposium on Microlithography, Paper 3049-19, March 9-14, 1997.
12. W.L. Krisa, "DUV Resist Processing at 0.25 ?m and below," SEMATECH DUV Lithography Workshop, Austin, TX, Oct. 16-18, 1996.
13. W.G. Oldham, R.E. Schenker, "193-nm Lithographic System Lifetimes as Limited by UV Compaction," Solid State Technology, pp. 95-102, April 1997.
14. C.R. Szmanda et al., "Resists for 193 nm Lithography: Physicochemical Influences on Resist Performance," SPIE`s 22nd International Symposium on Microlithography, Paper 3049-03, March 9-14, 1997.
15. F.M. Houlihan et al., "Recent Advances in 193 nm Single-layer Photoresists Based on Alternating Copolymer Cycloolefins," SPIE`s 22nd International Symposium on Microlithography, Paper 3049-05, March 9-14, 1997.
16. M.A. Piestrup, S.J. Mrowka, M.W. Powell, "Single-stepper Soft X-ray Source for Step-and-Scan Tools," SPIE`s 22nd International Symposium on Microlithography, Paper 3048-15, March 9-14, 1997.
17. J.H. Bruning, "Optical Lithography -Thirty Years and Three Orders of Magnitude," SPIE`s 22nd International Symposium on Microlithography, plenary session, March 9-14, 1997.
18. W.K. Waskiewicz et al., "SCALPEL Proof-of-Concept System: Preliminary Lithography Results," SPIE`s 22nd International Symposium on Microlithography, Paper 3048-23, March 9-14, 1997.
KATHERINE DERBYSHIRE is chief technical editor of Solid State Technology. She received her BS degree in materials science and engineering from the Massachusetts Institute of Technology, and her MS degree in engineering materials from the University of California, Santa Barbara. Before joining Solid State Technology as Senior Technical Editor, she had published research on high-temperature superconductors, III-V semiconductors, diamond thin films, and archaeological bronzes. Her current interests include lithography, thermal processing, and fab automation and management. Solid State Technology, 10 Tara Blvd., Fifth Floor, Nashua, NH 03062; ph 603/891-9216, e-mail [email protected].